Working alongside an AI-powered “partner” encourages women to participate

A virtual team member powered by artificial intelligence and speaking in a female voice increases women’s participation and productivity in male-dominated teams, according to a new Cornell study.

The findings suggest that the gender of an AI’s voice can positively impact the dynamics of gender-imbalanced teams and could aid in the design of bots used for human-AI teamwork, researchers say.

“I didn’t expect that having an AI agent on the team would replicate many of the same results that other researchers have observed with all-human, gender-unbalanced teams,” said Angel Hsing-Chi Hwang, a postdoctoral fellow in computer science at Cornell Ann S. Bowers College of Computer and Information Science. “It was a surprise to me.”

Hwang is the lead author of “The Sound of Support: Gendered Voice Agent as Support to Minority Teammates in Gender-Imbalanced Team,” which received honorable mention for the best paper award at the Association for Computing Machinery (ACM) CHI conference on Human Factors in Computing Systems, which took place on May 11–16.

The findings mirror previous research in psychology and organizational behavior that has shown that minority team members are more likely to participate if the team adds members who are similar to them, Hwang said.

“But hiring a new person to fill out your team in real time is unrealistic,” said Hwang, who studies the impact of artificial intelligence on work practices. “Our thought was: What if we had AI agents on demand that could participate and hopefully positively change the team dynamic?”

To better understand how artificial intelligence can help gender-imbalanced teams, Hwang and Andrea Stevenson Won, associate professor of communications in the College of Agriculture and Life Sciences and co-author of the paper, conducted an experiment with approximately 180 men and women. who were assigned to groups of three and asked to collaborate virtually on a set of tasks (only participants who identified as male or female were included in the study).

Each group consisted of one woman or one man and a fourth agent, an abstract shape with a male or female voice, who appeared on the screen and read instructions, submitted ideas, and kept time. There was a catch – the bot was not fully automated. In a process referred to in human-computer interaction as the “Wizard of Oz” experiment, Hwang was behind the scenes and fed lines generated by ChatGPT to the bot.

After the experiment, Hwang and Won analyzed team conversation logs to determine how often participants presented ideas or arguments. They also asked participants to reflect on the level of support offered, the experience of their team, and whether they personally felt marginalized, either by their human team members or by the bot.

“When we looked at the actual behavior of the participants, we started to see differences between men and women and how they reacted when there was a female agent on the team,” she said.

“The interesting thing about this study is that most participants did not prefer a male- or female-sounding voice,” Won said. “This means that people’s social inferences about AI can have an impact, even if people don’t believe they are important.”

Minority women were more likely to participate when the AI voice was female, while minority men were more talkative but less task-focused when working with a bot with a male voice, researchers found. According to the researchers, unlike men, women reported significantly more positive perceptions of an AI team member when they were in a minority.

“By using only a gender-sensitive voice, an AI agent can provide little support to minority members of a group,” said Hwang, who will join the faculty of the Annenberg School for Communication and Journalism at the University of Southern California this fall.

As for why, Hwang points to existing research on team dynamics.

“We often feel more comfortable and therefore work better in teams made up of people who are similar to us,” she said.

-Note: This press release was originally published on the Cornell University website. Because it has been republished, it may differ from our style guide.